Fault Data Mining Of Switch Machine Based On Clustering Ensemble

The switch machine is an important component of high-speed railway signal equipment whose normal operation is related to the safety of high-speed trains.S700 K switch machine is widely used in high-speed railway of China.It is a signal equipment with high probability of failure due to it is located outside.At present,the maintenance of the switch machine mostly depend on the maintainers' field experience and theoretical knowledge,which is carried out on a regular basis.So the judgment and processing efficiency of the switch machine's fault is limited.Clustering ensemble is the integration of multiple basis clustering results.It can improve generalization ability of the system and deal with the outlier better.As a result,clustering ensemble can improve the accuracy of the final clustering results.This thesis introduces clustering ensemble into the research of switch machine's fault data mining.Fault diagnosis of switch machine from the point of machine learning.The main work is as follows.Firstly,the monitoring power data curve of the switch machine is analyzed to show that the power data can reflect the working condition.According to the characteristics of switch machine's power monitoring data,clustering ensemble is carried out to mine the fault data.In data preprocessing phase,this thesis use segmentation method,AR model,PCA and DTW to solve the problem that the dimension of the power data is high and different.Also,obtain the common fault feature data set of the switch machine.Next,use K-means,FCM,agglomerative hierarchy clustering,DBSCAN and DPCA as the based clustering algorithms and match different data preprocessing methods with the based clustering algorithm to get the based clustering members and compare the accuracy of the based clustering results.Then get the adaptability of five clustering algorithms and four preprocessing methods for the power data of switch machine.And this is the basis of the optimization on based clustering members.At last,construct the general framework of the switch machine's fault data mining based on clustering ensemble.Design the W-VOTE ensemble method by a kind of weight voting based on the results of clustering.Then,complete the fault data mining by using W-VOTE,CSPA and MCLA to integrate the based clustering members.Compare the accuracy of ensemble methods and single clustering algorithms,and verify the superiority of clustering ensemble in fault data mining of switch machine.It helps to realize the active maintenance and reduce maintenance cost of the switch machine.